Multi-scale optimization method and system for optical storage charging station considering multiple uncertainties

Abstract

This invention proposes a multi-scale optimization method and system for photovoltaic-storage charging stations that considers multiple uncertainties. The method first considers the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices, constructing a day-ahead scheduling model and an intraday rolling optimization model for the photovoltaic-storage charging station. Then, it solves the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station. The day-ahead scheduling plan is then input into the intraday rolling optimization model to obtain the optimal decision for each rolling optimization period within the day. Finally, a real-time optimization model is constructed to obtain the real-time control commands for the photovoltaic-storage charging station system. This invention, through multi-timescale collaborative optimization at the day-ahead, intraday, and real-time scales, more intelligently plans the charging and discharging strategies for energy storage, avoiding ineffective cycles and excessive losses. It maximizes the local consumption and value creation of clean energy, effectively balancing the economy and robustness of the photovoltaic-storage charging station system, and enhancing the station's ability to cope with uncertainties.

Description

Technical Field

[0001] This invention belongs to the field of multi-timescale optimization of photovoltaic-storage-charging stations, specifically involving a multi-scale optimization method and system for photovoltaic-storage-charging stations that takes into account multiple uncertainties. Background Technology

[0002] With the acceleration of the global energy transition, the proportion of renewable energy, represented by photovoltaics, in the power system continues to increase. However, the intermittency and volatility of photovoltaic power generation, as well as the randomness of electric vehicle charging loads, pose serious challenges to the stable and economical operation of the distribution network, especially photovoltaic-storage-charging stations that integrate these elements. This multi-layered uncertainty makes traditional optimization scheduling methods based on deterministic predictions difficult to apply, often resulting in high operating costs or infeasible scheduling plans.

[0003] To address the multiple uncertainties inherent in photovoltaic-storage charging stations, existing technologies propose a stochastic optimization strategy for the daily operation of such stations for electric buses. A stochastic optimization strategy model for the photovoltaic-storage station is constructed, and during the day-ahead phase, the initial energy storage capacity is optimized across multiple scenarios with the goal of minimizing daily operating costs. However, this approach neglects the uncertainties in photovoltaic power generation within the station, as well as the uncertainties in electric vehicle scheduling. To address the high operating costs of photovoltaic-storage charging stations and the load fluctuations they cause to the distribution network due to insufficient operational strategies, a day-ahead game-theoretic economic scheduling method for the distribution network and photovoltaic-storage charging stations is proposed. This method considers the uncertainties in photovoltaic output and charging demand, employing Monte Carlo simulation and K-means++ clustering to generate and reduce photovoltaic output and electric vehicle charging scenarios. However, it does not consider multi-timescale scheduling and is inaccurate in local real-time control stages. Therefore, how to achieve coordinated optimization and scheduling of photovoltaic-storage charging stations to maximize overall system benefits has become a key research and practical issue. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned problems in the existing technology by providing a multi-scale optimization method and system for photovoltaic energy storage charging stations that takes into account multiple uncertainties.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] In a first aspect, this invention proposes a multi-scale optimization method for photovoltaic-storage-charging stations that takes into account multiple uncertainties, including:

[0007] S1. Considering the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices, construct a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage charging stations;

[0008] S2. Solve the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station;

[0009] S3. Input the day-ahead scheduling plan of the photovoltaic-storage charging station into the intraday rolling optimization model to obtain the optimal decision for each intraday rolling optimization period;

[0010] S4. Based on the optimal decision for each rolling optimization period within the day, a real-time layer optimization model is constructed to obtain the real-time control commands for the photovoltaic-storage charging station system.

[0011] S2 includes:

[0012] S21. Construct the global uncertainty vector of the photovoltaic-storage-charging station system:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] ;

[0018] In the above formula, This is the global uncertainty vector. For time period The uncertainty vector, This represents the total number of time periods within the scheduling cycle. For time period The error in photovoltaic power output prediction For time period Electric vehicle charging power prediction error, For time period The fluctuation deviation between the real-time electricity price and the planned electricity price before the day, For time period The actual power of photovoltaic power generation at that time For time period The predicted value of photovoltaic power generation, For time period The actual total power of electric vehicles For time period The total predicted power of electric vehicles, For time period Real-time electricity prices, including purchase price and sales price. For time period The planned electricity price for the day;

[0019] S22. Construct an adaptive Wasserstein fuzzy set based on the global uncertainty vector:

[0020] ;

[0021] ;

[0022] ;

[0023] In the above formula, For adaptive Wasserstein fuzzy sets, Global uncertainty vector The true probability distribution To support the collection The set of all probability distributions above, the support set Global uncertainty vector The set of all possible values. for and The 1-Wasserstein distance between two distributions is used to measure the difference between them. Distribution based on historical experience, For adaptive Wasserstein fuzzy set radius, This provides an estimate of the local kernel density of the current predicted scenario within historical scenarios. Based on the radius, For adaptive adjustment coefficients, For the number of historical scene samples, For kernel function, For the first A sample of historical scenes;

[0024] S23. Based on adaptive Wasserstein fuzzy sets, construct a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage-charging stations.

[0025] In S23, the day-ahead scheduling model of the photovoltaic-storage charging station is a two-stage sub-Bruker optimization model. The first stage refers to the day-ahead baseline plan that must be made before the uncertainty is realized, and the second stage refers to the real-time adjustment decision made based on the day-ahead baseline plan after the uncertainty is realized.

[0026] The first-stage optimization model for the split-bar is as follows:

[0027] ;

[0028] ;

[0029] ;

[0030] ;

[0031] In the above formula, for feasible domain, These are the decision variables for the first phase of the recent scheduling. The first-stage cost is calculated based on the planned values ​​of the decision variables. For grid interaction costs, For energy storage depreciation costs, This represents the total number of time periods within the scheduling cycle. , Time periods Electricity purchase price and electricity sales price , These are the time periods planned for the current day. The power purchased from the grid and the power sold to the grid at that time. For time interval step, The energy storage depreciation cost per unit charge / discharge capacity. , These are the time periods planned for the current day. The charging power and discharging power of the energy storage device;

[0032] The decision variables for the first stage are:

[0033] ;

[0034] In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy;

[0035] The constraints of the first-stage sub-Bruker optimization model include system power balance constraints, photovoltaic constraints, energy storage system constraints, electric vehicle charging load constraints, and grid interaction constraints.

[0036] The system power balance constraint is:

[0037] ;

[0038] In the above formula, For time period The actual power of photovoltaic power generation at that time For time period Basic load of charging stations For time period Total electric vehicle charging load;

[0039] Photovoltaic constraints are:

[0040] ;

[0041] In the above formula, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic output prediction is one of the uncertainties that the system needs to deal with;

[0042] The constraints of the energy storage system are:

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] ;

[0050] In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The rated power of the energy storage , These are binary variables representing energy storage charging and discharging, respectively.

[0051] The electric vehicle charging load constraint is:

[0052] ;

[0053] ;

[0054] ;

[0055] ;

[0056] In the above formula, , The first The time it takes for an electric vehicle to arrive at and leave the photovoltaic-storage charging station. For time period Time assigned to the first The charging power of an electric vehicle For the first The charging demand of a vehicle. For the first The maximum power of a charging station for electric vehicles. For time period The total power of the electric vehicle at that time For electric vehicle collection;

[0057] The power grid interaction constraints are:

[0058] ;

[0059] ;

[0060] ;

[0061] In the above formula, , These are binary variables for purchasing and selling electricity, respectively. This represents the maximum power output of the power grid.

[0062] The second-stage optimization model for the Brussels bar is as follows:

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] In the above formula, For the true probability distribution, For adaptive Wasserstein fuzzy sets, The value function for the second stage represents the value given the decision made in the first stage. Given the realization of uncertainty, finding the minimum cost required for optimal adjustment is itself an optimization problem. This represents the total number of time periods within the scheduling cycle. This is the cost coefficient for the second phase of adjustment. To adjust the amount, To account for the cost of energy storage battery degradation, , Time periods The charging power and discharging power of energy storage For time period Actual state of charge adjustment of energy storage It is a dynamic coefficient related to the SOC state. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For time interval step, The investment cost per unit capacity of energy storage batteries, The rated capacity of the battery. For the battery at the reference depth of discharge The number of loops below, For time period depth of discharge, Battery life degradation index For time period The initial state of charge of the stored energy;

[0069] The decision variables for the second stage are:

[0070] ;

[0071] In the above formula, For the decision variables of the second phase of the current scheduling, This is the global uncertainty vector. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Adjustment amount of electric vehicle charging load;

[0072] The constraints of the second-stage sub-Bluerg optimization model include power balance constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints.

[0073] The power balance constraint is:

[0074] ;

[0075] In the above formula, For time period Adjustment amount of power purchased from the grid, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic power output prediction For time period Basic load of charging stations For time period Total predicted electric vehicle power, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of power sold to the grid;

[0076] The actual charge / discharge constraint is:

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] In the above formula, The rated power of the energy storage , These are binary variables representing the actual charging and discharging of the energy storage, respectively, characterizing the actual operating state. It is a sufficiently large constant;

[0083] The actual constraints of the SOC are:

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] In the above formula, For time period The actual state of charge of the energy storage at the beginning. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For time period The actual charging power of energy storage For time interval step, For time period The actual discharge power of the energy storage For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively;

[0089] The power grid interaction constraints are as follows:

[0090] ;

[0091] ;

[0092] ;

[0093] ;

[0094] ;

[0095] In the above formula, This is the maximum power of the power grid. , These are binary variables representing the actual electricity purchases and sales, respectively.

[0096] In S23, the objective function of the intraday rolling optimization model is:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] In the above formula, Let be the decision variable, representing the time period determined in the intraday rolling optimization. The intraday planned value, Let be an optimized coverage window, representing the time set of the scheduling cycle for the intraday rolling optimization model. For operating costs, consistent with the current day's target, the calculation window uses the latest forecasted operating costs. For planning deviation penalties, This is a time-varying adaptive penalty coefficient, dynamically adjusted according to the uncertainty level of ultra-short-term forecasting. For time period The day-ahead scheduling plan is a known quantity. Used to measure the magnitude of deviation. For the true probability distribution, Represents a fuzzy set of intraday uncertainty. This is the second-order value function for the intraday phase, with the same meaning as the day-ahead scheduling problem, but the time range is limited to the intraday rolling window. For each scrolling optimization moment, For a fixed prediction window, Basic rated penalty coefficient, This is the gain coefficient. For time period Uncertain mapping function , , These are the weighting coefficients for different degrees of uncertainty. For time period Photovoltaic power output, For time period Electric vehicle charging power, For time period Real-time electricity price;

[0102] The constraint body of the intraday rolling optimization model includes power balance constraints, initial state constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints.

[0103] The power balance constraint is:

[0104] ;

[0105] In the above formula, , These are the time periods in the intraday plan. The power purchased from the grid and the power sold to the grid at that time. For time period Adjustment amount of power purchased from the grid, For ultra-short-term photovoltaic power forecasting, For time period The error in photovoltaic power output prediction , These are the time periods in the intraday plan. The charging power and discharging power of energy storage. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Basic load of charging stations For the latest ultra-short-term electric vehicle power forecast, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of electric vehicle charging load. For time period Adjustment amount of power sold to the grid;

[0106] The initial state constraints are:

[0107] ;

[0108] In the above formula, For time period The state of charge, For time period Measured value of state of charge;

[0109] The actual charge / discharge constraints, actual SOC constraints, and grid interaction constraints are the same as the second-stage constraints of the day-ahead scheduling model, and the planned and adjusted values ​​of all equipment must be satisfied.

[0110] In S4, the objective function of the real-time layer optimization model is:

[0111] ;

[0112] ;

[0113] ;

[0114] In the above formula, The optimization goal for the real-time layer is... To predict the total number of steps in real time, , , These are all weighting coefficients for tracking error, respectively measuring the importance of tracking energy storage power, grid power, and SOC trajectory. For the current control moment, , , They are respectively in Time prediction The actual values ​​of energy storage at any given time, the actual power interacting with the grid, and the actual state of charge of the energy storage are the optimization variables for the real-time layer. , , They are obtained by intraday rolling optimization. The reference values ​​for energy storage at any given time, the reference values ​​for power interacting with the grid, and the reference values ​​for the state of charge of energy storage are known quantities. The weighting coefficient for the increment of the control quantity. To control the increase in quantity, For model credibility weights, This is the predicted output of the LSTM network for the system state. For the actual system output, The real-time system efficiency output by LSTM. To predict the time-domain step size in real time, The total energy that can be stored or provided under rated conditions. Compensation residuals for LSTM predictions;

[0115] The constraints of the real-time layer optimization model include power constraints and equipment operation constraints.

[0116] Secondly, this invention proposes a multi-scale optimization system for photovoltaic-storage-charging stations that takes into account multiple uncertainties, including a model building module, a day-ahead scheduling model solving module, an intraday rolling optimization module, and a real-time layer optimization module.

[0117] The model building module is used to consider the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices to build a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage charging stations.

[0118] The day-ahead scheduling model solving module is used to solve the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station;

[0119] The intraday rolling optimization module is used to input the day-ahead scheduling plan of the photovoltaic-storage charging station into the intraday rolling optimization model to obtain the optimal decision for each rolling optimization period within the day;

[0120] The real-time layer optimization module is used to construct a real-time layer optimization model based on the optimal decision of each rolling optimization period within the day, and obtain real-time control commands for the photovoltaic-storage charging station system.

[0121] The model building module includes a global uncertainty vector building unit, an adaptive Wasserstein fuzzy set building unit, and a day-ahead and intraday model building unit.

[0122] The global uncertainty vector construction unit is used to construct the global uncertainty vector of the photovoltaic-storage-charging station system.

[0123] ;

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] In the above formula, This is the global uncertainty vector. For time period The uncertainty vector, This represents the total number of time periods within the scheduling cycle. For time period The error in photovoltaic power output prediction For time period Electric vehicle charging power prediction error, For time period The fluctuation deviation between the real-time electricity price and the planned electricity price before the day, For time period The actual power of photovoltaic power generation at that time For time period The predicted value of photovoltaic power generation, For time period The actual total power of electric vehicles For time period The total predicted power of electric vehicles, For time period Real-time electricity prices, including purchase price and sales price. For time period The planned electricity price for the day;

[0129] The adaptive Wasserstein fuzzy set building unit is used to construct adaptive Wasserstein fuzzy sets based on the global uncertainty vector:

[0130] ;

[0131] ;

[0132] ;

[0133] In the above formula, For adaptive Wasserstein fuzzy sets, Global uncertainty vector The true probability distribution To support the collection The set of all probability distributions above, the support set Global uncertainty vector The set of all possible values. for and The 1-Wasserstein distance between two distributions is used to measure the difference between them. Distribution based on historical experience, For adaptive Wasserstein fuzzy set radius, This provides an estimate of the local kernel density of the current predicted scenario within historical scenarios. Based on the radius, For adaptive adjustment coefficients, For the number of historical scene samples, For kernel function, For the first A sample of historical scenes;

[0134] The day-ahead and intraday model building units are used to construct day-ahead scheduling models and intraday rolling optimization models for photovoltaic-storage-charging stations based on adaptive Wasserstein fuzzy sets.

[0135] In the day-ahead and intraday model construction unit, the day-ahead scheduling model of the photovoltaic-storage charging station is a two-stage sub-Bruker optimization model. The first stage refers to the day-ahead baseline plan that must be made before the uncertainty is realized, and the second stage refers to the real-time adjustment decision made based on the day-ahead baseline plan after the uncertainty is realized.

[0136] The first-stage optimization model for the split-bar is as follows:

[0137] ;

[0138] ;

[0139] ;

[0140] ;

[0141] In the above formula, for feasible domain, These are the decision variables for the first phase of the recent scheduling. The first-stage cost is calculated based on the planned values ​​of the decision variables. For grid interaction costs, For energy storage depreciation costs, This represents the total number of time periods within the scheduling cycle. , Time periods Electricity purchase price and electricity sales price , These are the time periods planned for the current day. The power purchased from the grid and the power sold to the grid at that time. For time interval step, The energy storage depreciation cost per unit charge / discharge capacity. , These are the time periods planned for the current day. The charging power and discharging power of the energy storage device;

[0142] The decision variables for the first stage are:

[0143] ;

[0144] In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy;

[0145] The constraints of the first-stage sub-Bruker optimization model include system power balance constraints, photovoltaic constraints, energy storage system constraints, electric vehicle charging load constraints, and grid interaction constraints.

[0146] The system power balance constraint is:

[0147] ;

[0148] In the above formula, For time period The actual power of photovoltaic power generation at that time For time period Basic load of charging stations For time period Total electric vehicle charging load;

[0149] Photovoltaic constraints are:

[0150] ;

[0151] In the above formula, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic output prediction is one of the uncertainties that the system needs to deal with;

[0152] The constraints of the energy storage system are:

[0153] ;

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] ;

[0159] ;

[0160] In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The rated power of the energy storage , These are binary variables representing energy storage charging and discharging, respectively.

[0161] The electric vehicle charging load constraint is:

[0162] ;

[0163] ;

[0164] ;

[0165] ;

[0166] In the above formula, , The first The time it takes for an electric vehicle to arrive at and leave the photovoltaic-storage charging station. For time period Time assigned to the first The charging power of an electric vehicle For the first The charging demand of a vehicle. For the first The maximum power of a charging station for electric vehicles. For time period The total power of the electric vehicle at that time For electric vehicle collection;

[0167] The power grid interaction constraints are:

[0168] ;

[0169] ;

[0170] ;

[0171] In the above formula, , These are binary variables for purchasing and selling electricity, respectively. This represents the maximum power output of the power grid.

[0172] The second-stage optimization model for the Brussels bar is as follows:

[0173] ;

[0174] ;

[0175] ;

[0176] ;

[0177] ;

[0178] In the above formula, For the true probability distribution, For adaptive Wasserstein fuzzy sets, The value function for the second stage represents the value given the decision made in the first stage. Given the realization of uncertainty, finding the minimum cost required for optimal adjustment is itself an optimization problem. This represents the total number of time periods within the scheduling cycle. This is the cost coefficient for the second phase of adjustment. To adjust the amount, To account for the cost of energy storage battery degradation, , Time periods The charging power and discharging power of energy storage For time period Actual state of charge adjustment of energy storage It is a dynamic coefficient related to the SOC state. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For time interval step, The investment cost per unit capacity of energy storage batteries, The rated capacity of the battery. For the battery at the reference depth of discharge The number of loops below, For time period depth of discharge, Battery life degradation index For time period The initial state of charge of the stored energy;

[0179] The decision variables for the second stage are:

[0180] ;

[0181] In the above formula, For the decision variables of the second phase of the current scheduling, This is the global uncertainty vector. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Adjustment amount of electric vehicle charging load;

[0182] The constraints of the second-stage sub-Bluerg optimization model include power balance constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints.

[0183] The power balance constraint is:

[0184] ;

[0185] In the above formula, For time period Adjustment amount of power purchased from the grid, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic power output prediction For time period Basic load of charging stations For time period Total predicted electric vehicle power, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of power sold to the grid;

[0186] The actual charge / discharge constraint is:

[0187] ;

[0188] ;

[0189] ;

[0190] ;

[0191] ;

[0192] In the above formula, The rated power of the energy storage , These are binary variables representing the actual charging and discharging of the energy storage, respectively, characterizing the actual operating state. It is a sufficiently large constant;

[0193] The actual constraints of the SOC are:

[0194] ;

[0195] ;

[0196] ;

[0197] ;

[0198] In the above formula, For time period The actual state of charge of the energy storage at the beginning. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For time period The actual charging power of energy storage For time interval step, For time period The actual discharge power of the energy storage For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively;

[0199] The power grid interaction constraints are as follows:

[0200] ;

[0201] ;

[0202] ;

[0203] ;

[0204] ;

[0205] In the above formula, This is the maximum power of the power grid. , These are binary variables representing the actual electricity purchased and sold, respectively.

[0206] In the aforementioned day-ahead and intraday model construction units, the objective function of the intraday rolling optimization model is:

[0207] ;

[0208] ;

[0209] ;

[0210] ;

[0211] In the above formula, Let be the decision variable, representing the time period determined in the intraday rolling optimization. The intraday planned value, Let be an optimized coverage window, representing the time set of the scheduling cycle for the intraday rolling optimization model. For operating costs, consistent with the current day's target, the calculation window uses the latest forecasted operating costs. For planning deviation penalties, This is a time-varying adaptive penalty coefficient, dynamically adjusted according to the uncertainty level of ultra-short-term forecasting. For time period The day-ahead scheduling plan is a known quantity. Used to measure the magnitude of deviation. For the true probability distribution, Represents a fuzzy set of intraday uncertainty. This is the second-order value function for the intraday phase, with the same meaning as the day-ahead scheduling problem, but the time range is limited to the intraday rolling window. For each scrolling optimization moment, For a fixed prediction window, Basic rated penalty coefficient, This is the gain coefficient. For time period Uncertain mapping function , , These are the weighting coefficients for different degrees of uncertainty. For time period Photovoltaic power output, For time period Electric vehicle charging power, For time period Real-time electricity price;

[0212] The constraint body of the intraday rolling optimization model includes power balance constraints, initial state constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints.

[0213] The power balance constraint is:

[0214] ;

[0215] In the above formula, , These are the time periods in the intraday plan. The power purchased from the grid and the power sold to the grid at that time. For time period Adjustment amount of power purchased from the grid, For ultra-short-term photovoltaic power forecasting, For time period The error in photovoltaic power output prediction , These are the time periods in the intraday plan. The charging power and discharging power of energy storage. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Basic load of charging stations For the latest ultra-short-term electric vehicle power forecast, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of electric vehicle charging load. For time period Adjustment amount of power sold to the grid;

[0216] The initial state constraints are:

[0217] ;

[0218] In the above formula, For time period The state of charge, For time period Measured value of state of charge;

[0219] The actual charge / discharge constraints, actual SOC constraints, and grid interaction constraints are the same as the second-stage constraints of the day-ahead scheduling model, and the planned and adjusted values ​​of all equipment must be satisfied.

[0220] In the real-time layer optimization module, the objective function of the real-time layer optimization model is:

[0221] ;

[0222] ;

[0223] ;

[0224] In the above formula, The optimization goal for the real-time layer is... To predict the total number of steps in real time, , , These are all weighting coefficients for tracking error, respectively measuring the importance of tracking energy storage power, grid power, and SOC trajectory. For the current control moment, , , They are respectively in Time prediction The actual values ​​of energy storage at any given time, the actual power interacting with the grid, and the actual state of charge of the energy storage are the optimization variables for the real-time layer. , , They are obtained by intraday rolling optimization. The reference values ​​for energy storage at any given time, the reference values ​​for power interacting with the grid, and the reference values ​​for the state of charge of energy storage are known quantities. The weighting coefficient for the increment of the control quantity. To control the increase in quantity, For model credibility weights, This is the predicted output of the LSTM network for the system state. For the actual system output, The real-time system efficiency output by LSTM. To predict the time-domain step size in real time, The total energy that can be stored or provided under rated conditions. Compensation residuals for LSTM predictions;

[0225] The constraints of the real-time layer optimization model include power constraints and equipment operation constraints.

[0226] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0227] 1. This invention proposes a multi-scale optimization method and system for photovoltaic-storage charging stations that considers multiple uncertainties. The method first considers the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices, constructing a day-ahead scheduling model and an intraday rolling optimization model for the photovoltaic-storage charging station. Then, it solves the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station. The day-ahead scheduling plan is then input into the intraday rolling optimization model to obtain the optimal decision for each rolling optimization period within the day. Finally, based on the optimal decision for each rolling optimization period within the day, a real-time optimization model is constructed to obtain the real-time control commands for the photovoltaic-storage charging station system. This method, through multi-timescale collaborative optimization (day-ahead, intraday, real-time), more intelligently plans the charging and discharging strategies of energy storage, avoiding ineffective cycles and excessive losses. It maximizes the local consumption and value creation of clean energy, effectively balances the economy and robustness of the photovoltaic-storage charging station system, and enhances the station's ability to cope with uncertainties.

[0228] 2. This invention proposes a multi-scale optimization method and system for photovoltaic-storage-charging stations that takes into account multiple uncertainties. This method improves the dimension of uncertainty handling by considering the triple uncertainties of photovoltaic output, electric vehicle charging, and electricity price for coupled modeling. This enables more accurate capture of the operational risks of photovoltaic-storage-charging stations and introduces an adaptive adjustment mechanism to reduce the radius in dense sample areas to improve economic efficiency and increase the radius in sparse sample areas to enhance robustness. In the complex and ever-changing photovoltaic-storage-charging station system, this achieves dual optimization of safe operation and economic benefits.

[0229] 3. This invention proposes a multi-scale optimization method and system for photovoltaic-storage-charging stations that takes into account multiple uncertainties. In the intraday rolling optimization, this method introduces a planning deviation penalty term, dynamically adjusts the time-varying adaptive penalty coefficient based on the level of uncertainty in ultra-short-term predictions, enhances the coordination between intraday plans and day-ahead plans through mathematical constraints, and effectively smooths out power fluctuations in ultra-short time scales through continuous rolling optimization, thereby achieving short-term economy, coordination with day-ahead plans, and robustness.

[0230] 4. This invention proposes a multi-scale optimization method and system for photovoltaic-storage-charging stations that takes into account multiple uncertainties. This method considers the nonlinear efficiency error existing in the photovoltaic-storage-charging station system, constructs a real-time layer optimization model to provide feedback adjustment for the day-ahead and intraday scheduling, ensures the smoothness of control actions, reduces the impact of nonlinear efficiency error, accurately tracks upper-level instructions, and ensures that the actual operating trajectory of the system closely follows the economically optimal path formulated by the optimization layer. Attached Figure Description

[0231] Figure 1 This is an overall flowchart of the method described in this invention.

[0232] Figure 2 This is a diagram showing the scheduling results of the photovoltaic-storage charging station described in Example 1.

[0233] Figure 3 This is a structural diagram of the system described in this invention. Detailed Implementation

[0234] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0235] This invention proposes a multi-scale optimization method and system for photovoltaic-storage-charging stations that considers multiple uncertainties. It constructs a system model of the photovoltaic-storage-charging station, including photovoltaic power generation units, energy storage battery units, and electric vehicle charging loads. Then, it constructs an adaptive Wasserstein fuzzy set with multi-dimensional coupling of "source-load-price" to handle the triple uncertainties of photovoltaic output, charging demand, and electricity price. Next, it uses a data-driven sub-Benders decomposition optimization method to formulate a day-ahead scheduling plan, solving it using an adaptive scenario generation and multi-cut Benders decomposition collaborative optimization algorithm, and performing rolling optimization in the intraday stage, using a rolling time domain combined with linear programming (LP) to solve the problem, dynamically adjusting the scheduling plan based on the improvement of prediction accuracy. Finally, in the real-time stage, it uses a method based on fused LSTM-MPC to predict and correct nonlinear efficiency errors caused by energy storage battery aging, environmental temperature changes, etc. Through multi-timescale collaborative optimization of "day-ahead-intraday-real-time," it effectively balances the economy and robustness of system operation, improving the photovoltaic-storage-charging station's ability to cope with uncertainties.

[0236] Example 1:

[0237] like Figure 1 As shown, the multi-scale optimization method for photovoltaic-storage-charging stations, which takes into account multiple uncertainties, is carried out in the following steps:

[0238] 1. Construct an optimization model for the photovoltaic-storage-charging station system;

[0239] The photovoltaic-storage charging station system mainly consists of four parts: photovoltaic power generation unit, energy storage battery unit, electric vehicle charging load, and interface with the power distribution network;

[0240] Initialize the scheduling cycle time set for establishing the optimization model of the photovoltaic-storage-charging station system. and electric vehicle clusters ,in, This represents the total number of time periods within the scheduling cycle. Indicates the time period A collection of electric vehicles located within the station and capable of being dispatched;

[0241] The objective function of the photovoltaic-storage-charging station system optimization model, with the goal of minimizing the total operating cost of the system within a scheduling cycle, is as follows:

[0242] ;

[0243] ;

[0244] ;

[0245] In the above formula, The total operating cost of the photovoltaic-storage charging station system, For grid interaction costs, For energy storage depreciation costs, This represents the total number of time periods within the scheduling cycle. , Time periods Electricity purchase price and electricity sales price , Time periods Power purchased from the grid and power sold to the grid. For time interval step, The energy storage depreciation cost per unit charge / discharge capacity. , Time periods The charging power and discharging power of energy storage.

[0246] The constraints of the photovoltaic-storage charging station system optimization model include system power balance constraints, photovoltaic constraints, energy storage system constraints, electric vehicle charging load constraints, and grid interaction constraints.

[0247] Among them, for all time periods The system power balance constraint is:

[0248] ;

[0249] In the above formula, , Time periods Power purchased from the grid and power sold to the grid. For time period The actual power of photovoltaic power generation at that time , Time periods The charging power and discharging power of energy storage For time period The basic load of the charging station (such as office power consumption, lighting, etc.) is a known quantity. For time period Total electric vehicle charging load;

[0250] Because photovoltaic output is uncertain, it is necessary to distinguish between predicted and actual values ​​in the modeling process. The photovoltaic constraints are as follows:

[0251] ;

[0252] In the above formula, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic output prediction is one of the uncertainties that the system needs to deal with;

[0253] The energy storage system is the core controllable unit of a photovoltaic-energy storage charging station. It requires precise description of energy dynamics and operational limitations. The constraints of the constructed energy storage system are as follows:

[0254] ;

[0255] ;

[0256] ;

[0257] ;

[0258] ;

[0259] ;

[0260] ;

[0261] In the above formula, For time period The initial state of charge of the stored energy. , The charging efficiency and discharging efficiency of energy storage respectively meet the requirements. , For time period The charging power of energy storage, and , For time interval step, For time period The discharge power of the stored energy, and , For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The rated power of the energy storage , These are binary variables for charging and discharging energy storage, respectively, to prevent simultaneous charging and discharging. ;

[0262] Electric vehicle charging load is another major uncertainty and schedulable resource. Electric vehicle charging load constraints include charging demand constraints, power upper and lower limit constraints, and aggregate load constraints.

[0263] The charging demand constraint is:

[0264] ;

[0265] The upper and lower limits of power are constrained as follows:

[0266] ;

[0267] ;

[0268] The aggregate load constraint is:

[0269] ;

[0270] In the above formula, , The first The time it takes for an electric vehicle to arrive at and leave the photovoltaic-storage charging station. For time period Time assigned to the first The charging power of an electric vehicle For the first The charging demand of a vehicle. For the first The maximum power of a charging station for electric vehicles. For time period The total power of an electric vehicle is the optimal decision value for electric vehicles, which is currently obtained through prediction. After implementing the method proposed in this paper, the optimized scheduling value is obtained. For electric vehicle collection;

[0271] The photovoltaic-storage charging station exchanges energy with the distribution network through a common connection point. The grid interaction model constructed is as follows:

[0272] ;

[0273] ;

[0274] ;

[0275] In the above formula, For time period Power purchased from the grid, and , , These are binary variables representing electricity purchase and electricity sales, respectively. , This is the maximum power of the power grid. For time period The power sold to the grid, and .

[0276] 2. Considering the uncertainties of photovoltaic power output, electric vehicle charging, and electricity prices, an adaptive Wasserstein fuzzy set with multidimensional coupling of "source-load-price" is constructed to handle the uncertainties of the photovoltaic-storage-charging station system optimization model;

[0277] The optimization model for a photovoltaic-storage charging station system includes three uncertainties: photovoltaic output, electric vehicle charging load demand, and time-of-use electricity price fluctuations. Modeling these uncertainties involves the following steps:

[0278] Based on the uncertainty parameters of the photovoltaic-storage-charging station system optimization model, a global uncertainty vector is constructed:

[0279] ;

[0280] ;

[0281] ;

[0282] ;

[0283] ;

[0284] In the above formula, This is the global uncertainty vector. For time period The uncertainty vector, This represents the total number of time periods within the scheduling cycle. For time period The error in photovoltaic power output prediction For time period The actual power of photovoltaic power generation at that time For time period The predicted value of photovoltaic power generation, For time period Electric vehicle charging power prediction error, For time period Total actual power of electric vehicles For time period Total predicted electric vehicle power, For time period The fluctuation deviation between the real-time electricity price and the planned electricity price before the day, For time period Real-time electricity prices, including electricity purchase price Electricity sales price , For time period The planned electricity price for the day;

[0285] In this invention, , Adjustments to these two variables will be reflected in adjustments to physical quantities, meaning their adjustments will affect power balance. Therefore, they must appear in the model as random variables, including electricity price errors. The adjustment mainly affects the cost coefficient in the objective function, thus reflecting the adjustment of electricity price error in the cost.

[0286] The present invention aims to perform robust optimization. The fluctuations of the motor are reflected in each historical sample. When solving for the optimal result, the cost of various combinations of photovoltaic, electric vehicles and corresponding electricity price scenarios will be calculated. Therefore, the adjustment of electricity price fluctuations is reflected in the objective function and the final calculated cost value.

[0287] Because the global uncertainty vector cannot be obtained. True probability distribution The present invention through A historical scene sample based on historical data , constitute the empirical distribution ,but It is based on limited data and is different from the true probability distribution. There will be deviations. Therefore, to prevent such deviations, a system is constructed based on... Unlike conventional methods that set the radius of the fuzzy set to a fixed value, this invention introduces an adaptive radius mechanism based on the local density of historical samples. The radius is reduced in dense sample regions to improve economy, and increased in sparse sample regions to enhance robustness. The specific formula is as follows:

[0288] ;

[0289] ;

[0290] ;

[0291] In the above formula, For adaptive Wasserstein fuzzy sets, For the true probability distribution, To support the collection The set of all probability distributions above, the support set For the studied photovoltaic-storage-charging station system, the global uncertainty vector is... The set of all possible values, that is, the entire range or space of all possible values ​​that an uncertain variable can take. for and The 1-Wasserstein distance between two distributions is used to measure the difference between them. Distribution based on historical experience, For adaptive Wasserstein fuzzy set radius, This provides an estimate of the local kernel density of the current predicted scenario within historical scenarios. Based on the radius, For adaptive adjustment coefficients, For the number of historical scene samples, For kernel function, For the first A sample of historical scenes.

[0292] 3. Based on adaptive Wasserstein fuzzy sets, construct a day-ahead scheduling model for photovoltaic-storage-charging stations;

[0293] The scheduling problem has recently been modeled as a two-stage optimization problem, where the first stage refers to the period before the uncertainty is realized, i.e., the prediction error is unknown. , , When determining specific values, the baseline plan must be made; this is the plan for the current day and cannot be easily changed once it is established. The second stage refers to the real-time adjustment decisions made based on actual photovoltaic output and electric vehicle load after the uncertainty materializes. The two-stage approach, by establishing the baseline in the first stage and making corrections in the second, maximizes operational economy while ensuring the plan's robustness to uncertainty, achieving the "optimal trade-off under the worst-case scenario." This solves the problem that single-stage robust optimization, in order to cope with extreme uncertainty, can be overly conservative, leading to poor scheduling plan economy.

[0294] First, determine the decision variables for the day-ahead scheduling problem:

[0295] The decision variables for the first stage are:

[0296] ;

[0297] In the above formula, These are the decision variables for the first phase of the recent scheduling. , These are the time periods planned for the current day. The charging power and discharging power of energy storage. , These are the time periods planned for the current day. The power purchased from the grid and the power sold to the grid at that time. For the planned time period in the previous day The initial state of charge of the stored energy;

[0298] The decision variables in the second stage are adjustments used to balance the random power bias, including:

[0299] ;

[0300] In the above formula, For the decision variables of the second phase of the current scheduling, This is the global uncertainty vector. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Adjustment amount of electric vehicle charging load;

[0301] Then, with the goal of minimizing the expected values ​​of the first-stage cost and the second-stage adjustment cost under the worst-case probability distribution, a two-stage sub-Bruker optimization model for the day-ahead scheduling problem is constructed.

[0302] The first-stage optimization model for the split-bar is as follows:

[0303] ;

[0304] ;

[0305] ;

[0306] ;

[0307] In the above formula, for feasible domain, These are the decision variables for the first phase of the recent scheduling. The first-stage cost is calculated based on the planned values ​​of the decision variables. For grid interaction costs, For energy storage depreciation costs, This represents the total number of time periods within the scheduling cycle. , Time periods Electricity purchase price and electricity sales price , These are the time periods planned for the current day. The power purchased from the grid and the power sold to the grid at that time. For time interval step, The energy storage depreciation cost per unit charge / discharge capacity. , These are the time periods planned for the current day. The charging power and discharging power of the energy storage device;

[0308] The constraints of the first-stage split-bar optimization model only involve planning and decision variables. The constraints are consistent with those of the photovoltaic-storage charging station system optimization model (the relevant variables are calculated using the planned values ​​of the decision variables), including system power balance constraints, photovoltaic constraints, energy storage system constraints, electric vehicle charging load constraints, and grid interaction constraints.

[0309] The second-stage optimization model for the split-bar is as follows:

[0310] ;

[0311] ;

[0312] ;

[0313] ;

[0314] ;

[0315] In the above formula, For the true probability distribution, For adaptive Wasserstein fuzzy sets, The value function for the second stage represents the value given the decision made in the first stage. Given the realization of uncertainty, finding the minimum cost required for optimal adjustment is itself an optimization problem. This represents the total number of time periods within the scheduling cycle. This is the cost coefficient for the second phase of adjustment. To adjust the amount, The cost of energy storage battery degradation, calculated based on actual charging and discharging decision variables, can be understood as "unit price × quantity". , Time periods The charging power and discharging power of energy storage For time period Actual state of charge adjustment of energy storage This is a dynamic coefficient related to the SOC state, which can be understood as the "unit price". , These are the charging efficiency and discharging efficiency of energy storage, respectively. For time interval step, The investment cost per unit capacity of energy storage batteries, The rated capacity of the battery. For the battery at the reference depth of discharge The number of loops below, For time period depth of discharge, Battery life degradation index For time period The initial state of charge of the stored energy;

[0316] The second stage of the sub-Blubar optimization model involves the design adjustment of constraints. Under the constraints, the actual operating power (planned value + adjusted value) of all equipment must meet physical constraints, including power balance constraints, actual charging and discharging constraints, actual SOC constraints, and grid interaction constraints.

[0317] The power balance constraint is the most critical constraint and must be applied to the support set. Almost all uncertainties within the constraint are realized (a,s), which links the planned value to the real-time adjusted value, ensuring the real-time power balance of the system under any circumstances:

[0318] ;

[0319] In the above formula, For time period Adjustment amount of power purchased from the grid, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic power output prediction For time period Basic load of charging stations For time period Total predicted electric vehicle power, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of power sold to the grid;

[0320] The actual charge / discharge constraint is:

[0321] ;

[0322] ;

[0323] ;

[0324] ;

[0325] ;

[0326] In the above formula, The rated power of the energy storage , These are binary variables representing the actual charging and discharging of the energy storage, respectively, characterizing the actual operating state. It is a sufficiently large constant;

[0327] The actual constraints of the SOC are:

[0328] ;

[0329] ;

[0330] ;

[0331] ;

[0332] In the above formula, For time period The actual state of charge of the energy storage at the beginning. , These are the charging efficiency and discharging efficiency of energy storage, respectively. For time period The actual charging power of energy storage For time interval step, For time period The actual discharge power of the energy storage For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively;

[0333] The power grid interaction constraints are as follows:

[0334] ;

[0335] ;

[0336] ;

[0337] ;

[0338] ;

[0339] In the above formula, This is the maximum power of the power grid. , These are binary variables representing the actual electricity purchases and sales, respectively.

[0340] 4. Solve the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station;

[0341] A two-stage Benders robust optimization model is used to solve the day-ahead scheduling problem using an adaptive scene generation and multi-cut Benders decomposition collaborative optimization algorithm, outputting a robust day-ahead schedule. This plan will serve as a benchmark for intraday rolling optimization;

[0342] Considering the high computational cost and slow convergence speed when processing large-scale scene sets, Benders decomposition is adopted to dynamically identify and optimize key scenes, generating independent Benders cuts for each key scene. This achieves more robust and faster convergence with less computation. The specific details are as follows:

[0343] The two-stage partial Blue Bar optimization of the current scheduling problem can be expressed as:

[0344] ;

[0345] ;

[0346] During the solution process, the strategy will be... With adjustment amount Use formula In summary, the value function in the second stage Defined as:

[0347] ;

[0348] ;

[0349] in, This reflects the linear impact of uncertainty on constraints;

[0350] Input historical scene samples Initialize the active scene set (Usually a small, randomly selected subset), fuzzy set radius Convergence tolerance Initial lower bound Initial upper bound Iteration counter ;

[0351] To accelerate convergence and more accurately approximate the original problem, the solution... Each cut retains a separate Benders cut, which includes the cost value for that scenario. and gradient information (Lagrange multipliers from subproblems) contain more information than a single cut;

[0352] In iteration Solve the following based on the current active scene set. The relaxation principal problem is solved to obtain the current solution. and target value (Lower bound of the original problem):

[0353] ;

[0354] ;

[0355] In the above formula, This is a dual variable introduced in the relaxation master problem (MP). Introducing auxiliary decision variables into the relaxation master problem (MP), For adaptive Wasserstein fuzzy set radius, For the first The set of active scenarios in the next iteration For the first Solution to the main problem in the next iteration. For the first A historical scene sample, For the first Gradient information of each historical scene, In order to be with the first A constant related to the historical scene and Wasserstein distance;

[0356] For all In each historical scenario, the second-order subproblem is solved in parallel, and the optimal value is recorded. and corresponding constraints The optimal Lagrange multiplier ,in, It is a constraint on the "price" or "sensitivity" at the optimal solution of the subproblem. Its core algorithmic value lies in providing a value function. exist This point is about Gradient information:

[0357] ;

[0358] ;

[0359] In the above formula, For the first Two-stage decision variables for a historical scenario For the first time planned The first-stage decision coefficient matrix of the next iteration Two-stage decision variables Technical coefficient matrix of system constraints, Let the vector be the constant terms of the equality constraints. This is the matrix of influence coefficients for uncertainty. This represents the feasible region for two-stage decision-making.

[0360] Calculate the inverse degree of each scene to achieve adaptive key scene generation:

[0361] ;

[0362] In the above formula, For the first Sub-iteration scenario The degree of rebellion quantifies the first In a specific historical scenario, the actual adjustment cost Beyond the dual variables and The degree to which the security boundary is constituted;

[0363] The greater the violation of a scenario, the higher the risk to the current solution, and the more "critical" the scenario needs to be. To ensure that optimization resources are concentrated on the few critical scenarios most likely to affect robustness, improve computational efficiency, and update the active scenario set, :

[0364] a. Add the K scenarios with the highest current inconsistency to the new active set.

[0365] b. Randomly select M new scenes from all N scenes to add in order to avoid local optima.

[0366] c. Remove scenarios with very low inverse behavior in consecutive iterations.

[0367] The steps are as follows: a) adding the most important new scene; b) introducing random exploration; and c) eliminating unimportant old scenes. These three steps are performed in sequence.

[0368] Calculate the current solution A robust upper bound valuation, updating the upper bound:

[0369] ;

[0370] judge If the condition is met, the algorithm converges and outputs the optimal solution. If not, then let And iteratively solve based on the updated active scene set.

[0371] The current plan is to use an adaptive scene generation and multi-cut Bender decomposition collaborative optimization algorithm to dynamically identify key scenes and accelerate convergence by generating independent cuts for each active scene, thereby obtaining more robust results with less computation.

[0372] 5. Construct an intraday rolling optimization model for photovoltaic-storage charging stations. Input the daytime scheduling plan of the photovoltaic-storage charging stations into the intraday rolling optimization model to obtain the optimal decision for each rolling optimization period within the day.

[0373] The intraday rolling optimization model is a two-endpoint blue bar optimization framework, as detailed below:

[0374] An adaptive planning deviation penalty term is introduced to construct the objective function of the intraday rolling optimization model. The objective function of intraday rolling optimization is a multi-objective complex, aiming to simultaneously achieve short-term economic efficiency, coordination with the day-ahead plan, and robustness. The scheduling plan is dynamically adjusted based on the improvement of forecast accuracy.

[0375] ;

[0376] ;

[0377] ;

[0378] ;

[0379] In the above formula, Let be the decision variable, representing the time period determined in the intraday rolling optimization. The intraday planned value, Let be an optimized coverage window, representing the time set of the scheduling cycle for the intraday rolling optimization model. For operating costs, consistent with the current day's target, the calculation window uses the latest forecasted operating costs. For planning deviation penalties, This is a time-varying adaptive penalty coefficient, dynamically adjusted according to the uncertainty level of ultra-short-term forecasting. For time period The day-ahead scheduling plan is a known quantity. Used to measure the magnitude of deviation. For the true probability distribution, Represents the intraday uncertainty fuzzy set, and Similarly, you only need to... The timescale can be changed to intraday. This is the second-order value function for the intraday phase, with the same meaning as the day-ahead scheduling problem, but the time range is limited to the intraday rolling window. For each scrolling optimization moment, For a fixed prediction window, Basic rated penalty coefficient, This is the gain coefficient, used to adjust the degree of influence of uncertainty on the penalty coefficient. For time period The mapping function of uncertainty is a mapping index for different levels of uncertainty. , , These are the weighting coefficients for different degrees of uncertainty. For time period Photovoltaic power output, For time period Electric vehicle charging power, For time period Real-time electricity price;

[0380] The constraint system of the intraday rolling optimization model has been updated and strengthened based on the day-ahead scheduling problem, including power balance constraints, initial state constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints.

[0381] The power balance constraint is:

[0382] ;

[0383] In the above formula, , These are the time periods in the intraday plan. The power purchased from the grid and the power sold to the grid at that time. For time period Adjustment amount of power purchased from the grid, For ultra-short-term photovoltaic power forecasting, For time period The error in photovoltaic power output prediction , These are the time periods in the intraday plan. The charging power and discharging power of energy storage. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Basic load of charging stations For the latest ultra-short-term electric vehicle power forecast, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of electric vehicle charging load. For time period Adjustment amount of power sold to the grid;

[0384] Initial state constraints force the initial state of the optimization problem to align with the actual measured values ​​of the system, achieving crucial feedback correction and avoiding error accumulation.

[0385] ;

[0386] In the above formula, For a moment The state of charge, These are measured values ​​of the state of charge.

[0387] The actual charge / discharge constraints, actual SOC constraints, and grid interaction constraints are the same as the second-stage constraints of the day-ahead scheduling model. The actual operating points (planned values ​​+ adjusted values) of all equipment must meet the physical and safety constraints, and in Pa.s. form. However, it should be noted that the dynamic evolution of energy storage SOC starts from the initial measured values ​​of the rolling plan.

[0388] To ensure real-time performance while enhancing the intelligence and robustness of optimization, Rolling Time Control (RHC) is combined with Linear Programming (LP). The intraday rolling optimization model is solved using this combined approach, outputting the results for each rolling optimization time step. Optimal decision The plan will be issued and implemented immediately. At the same time, the plan for subsequent time periods within the window will be reserved as a preliminary arrangement for further revision during the next rolling optimization. The specific details are as follows:

[0389] Initialization: Set the scroll window length , scrolling interval At present Enter the plan for the day. and adaptive penalty coefficient Adjustment rules;

[0390] State Measurement and Prediction Update: Acquisition Wait for real-time data and update the window. Short-term forecasts within the current period;

[0391] Adaptive parameter adjustment: dynamically calculated based on the latest prediction uncertainty assessment results. ;

[0392] Construct and solve the LP: Substitute the updated data, state, and adaptive parameters into the intraday rolling optimization model and solve it using an efficient LP solver (such as the simplex method);

[0393] To make the objective function of the intraday rolling optimization model more suitable for LP solution, it is necessary to adjust the nonlinear L1 norm. To achieve exact linearization, an auxiliary variable is introduced. Each decision variable component is processed as follows:

[0394] ;

[0395] ;

[0396] The final linear programming model is:

[0397] ;

[0398] ;

[0399] Obtain the optimal solution Only execute the current time. Instructions ;

[0400] Time scrolling: Return the state measurement and prediction update, and perform rolling optimization for the next time step.

[0401] 6. Based on the optimal decision of each rolling optimization period within the day, a real-time layer optimization model is constructed. The LSTM-MPC fusion method is used to predict and correct the nonlinear efficiency error in real time, and obtain the real-time control command of the photovoltaic-storage charging station system.

[0402] In the photovoltaic-storage charging station system, there are nonlinear efficiency errors caused by factors such as energy storage battery aging and changes in ambient temperature. Real-time optimization can accurately track upper-level instructions and is the most refined feedback control among the three stages. The purpose of real-time control is to provide feedback adjustment for the scheduling before and during the day, correct deviations to ensure the smoothness of control actions, and reduce the impact of nonlinear efficiency errors.

[0403] The objective function for constructing the real-time layer optimization model is:

[0404]

[0405] ;

[0406] In the original MPC state transition equation, the evolution of energy storage SOC is usually considered linear. This invention introduces an LSTM-based modified SOC prediction model, which is as follows:

[0407] ;

[0408] In the above formula, The optimization goal for the real-time layer is... To predict the total number of steps in real time, , , These are all weighting coefficients for tracking error, respectively measuring the importance of tracking energy storage power, grid power, and SOC trajectory. For the current control moment, , , They are respectively in Time prediction The actual values ​​of energy storage at any given time, the actual power interacting with the grid, and the actual state of charge of the energy storage are the optimization variables for the real-time layer. , , They are obtained by intraday rolling optimization. The reference values ​​for energy storage at any given time, the reference values ​​for power interacting with the grid, and the reference values ​​for the state of charge of energy storage are known quantities. The weighting coefficient for the increment of the control quantity. To control the increment, minimizing this term is to ensure smooth control actions and avoid impacting the energy storage device. This is a weight for model reliability. When the LSTM training error is large, this weight is reduced, degenerating into traditional MPC; when the LSTM accuracy is high, this weight is increased to achieve accurate tracking. The output of the LSTM network is a prediction of system states (such as power loss caused by voltage and temperature). For the actual system output, To improve the real-time system efficiency of LSTM. To predict the time-domain step size in real time, The total energy that can be stored or provided under rated conditions. Compensation residuals for LSTM predictions;

[0409] The constraints of the real-time layer optimization model include power constraints and equipment operation constraints;

[0410] The power constraint is:

[0411] ;

[0412] ;

[0413] In the above formula, for The actual measured values ​​of photovoltaic power at any given time demonstrate the feedforward compensation mechanism of MPC, which can immediately respond to and offset measurable real-time power disturbances. , They are respectively The actual charging and discharging power of the energy storage at all times. for The basic load of the charging station at all times. for The actual total measured value of photovoltaic at any given time. for The minimum charging power for continuous energy storage. This is the power loss factor, calculated by LSTM based on the current battery health and real-time temperature, to prevent forced high-power charging when the battery is at high temperature or extremely low SOC. The rated power of the energy storage;

[0414] The equipment operating constraints are:

[0415] ;

[0416] ;

[0417] ;

[0418] ;

[0419] ;

[0420] ;

[0421] ;

[0422] In the above formula, In order to be in Time prediction Actual values ​​of energy storage charging / discharging at any given time. To maximize the ramping power of energy storage, , They are respectively in Time prediction The actual value of power purchased from the grid and the actual value of power sold to the grid at any given time. To maximize energy storage capacity, In order to be in Time prediction Time assigned to the first The actual charging power of an electric vehicle , For the first Minimum and maximum power of electric vehicle charging stations This is the predicted value of the energy storage SOC. for Actual measured value of the storage state of energy (SOC) at any given time.

[0423] Solving the real-time layer optimization model includes:

[0424] The charging and discharging power, voltage, temperature, and SOC changes at different time periods are collected in real time and used as the input sequence for LSTM.

[0425] LSTM network predicts actual charge / discharge efficiency coefficients and nonlinear loss errors under current conditions. ;

[0426] The nonlinear parameters predicted by LSTM are injected into the linear predictor of MPC in real time to complete the adaptive correction of the model.

[0427] At each control moment The quadratic programming (QP) problem is solved based on the corrected high-precision model.

[0428] The optimal control sequence is obtained by solving the problem using a solver. Only real-time control commands It is then distributed to actuators such as energy storage converters.

[0429] Through continuous rolling optimization, MPC can effectively smooth out power fluctuations on ultra-short timescales, ensuring that the actual operating trajectory of the system closely follows the economically optimal path defined by the optimization layer.

[0430] The scheduling results of photovoltaic and energy storage charging stations are as follows Figure 2 As shown. Figure 2 In this process, the system operation is closely coordinated with the characteristics of photovoltaic power output: during the day when there is sufficient sunlight, photovoltaics become the main power source, and surplus electricity is intelligently allocated to meet the load within the station, charge energy storage, and sell electricity to the grid, maximizing the local consumption and value creation of clean energy. Secondly, as evening approaches and photovoltaic power declines while charging demand peaks, the key role of the optimization strategy becomes apparent. The energy storage system, based on optimization instructions, stores and discharges energy, working together with grid-purchased electricity to support the load. This "low-storage, high-output" mode effectively smooths out supply and demand fluctuations and significantly reduces the cost of purchasing electricity during periods of high electricity prices. The entire dispatching process demonstrates the friendly interaction between the system as a whole and the grid, forming an economical and reliable two-way energy flow. The dispatching results fully demonstrate the superior ability of this invention to cope with multiple uncertainties through multi-level optimization at the "day-ahead, intraday, and real-time" stages.

[0431] To verify the effectiveness of this scheme, it is compared with different scheduling schemes. The costs of each scheduling scheme are shown in Table 1. Scheme 1 is a traditional single-stage scheduling scheme, Scheme 2 is an improved three-stage scheduling scheme, and Scheme 3 is the scheduling scheme proposed in this invention. The improved three-stage scheduling scheme refers to the day-period-real-time three-stage scheduling scheme. The difference between this scheme and the one mentioned above is that this scheme is a three-stage scheme that considers demultiplexing optimization.

[0432] Table 1. Cost Comparison of Different Scheduling Methods

[0433] As can be seen from the data comparison in Table 1, the total cost of Scheme 3 (8999.9) is significantly lower than that of the improved three-stage scheduling of Scheme 2 (9792.3) and the traditional single-stage scheduling of Scheme 1 (11230), achieving a cost saving of over 20%. This indicates that the overall optimized architecture proposed in this scheme operates more economically. Furthermore, Scheme 3 shows the most significant reductions in energy storage cost (2108.2) and EV scheduling cost (3421.6). This is because this scheme, through precise multi-timescale optimization, more intelligently plans the charging and discharging strategies of energy storage, avoiding ineffective cycles and excessive losses, thereby significantly reducing energy storage depreciation costs. In summary, the scheme proposed in this invention demonstrates superior economic performance in photovoltaic-energy storage charging stations.

[0434] Example 2:

[0435] like Figure 3 As shown, the multi-scale optimization system for photovoltaic-storage-charging stations that takes into account multiple uncertainties includes a model building module, a day-ahead scheduling model solving module, an intraday rolling optimization module, and a real-time layer optimization module.

[0436] The model building module is used to consider the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices to build a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage charging stations.

[0437] The day-ahead scheduling model solving module is used to solve the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station;

[0438] The intraday rolling optimization module is used to input the day-ahead scheduling plan of the photovoltaic-storage charging station into the intraday rolling optimization model to obtain the optimal decision for each rolling optimization period within the day;

[0439] The real-time layer optimization module is used to construct a real-time layer optimization model based on the optimal decision of each rolling optimization period within the day, and obtain real-time control commands for the photovoltaic-storage charging station system.

[0440] The model building module includes a global uncertainty vector building unit, an adaptive Wasserstein fuzzy set building unit, and a day-ahead and intraday model building unit.

[0441] The global uncertainty vector construction unit is used to construct the global uncertainty vector of the photovoltaic-storage-charging station system.

[0442] ;

[0443] ;

[0444] ;

[0445] ;

[0446] ;

[0447] In the above formula, This is the global uncertainty vector. For time period The uncertainty vector, This represents the total number of time periods within the scheduling cycle. For time period The error in photovoltaic power output prediction For time period Electric vehicle charging power prediction error, For time period The fluctuation deviation between the real-time electricity price and the planned electricity price before the day, For time period The actual power of photovoltaic power generation at that time For time period The predicted value of photovoltaic power generation, For time period The actual total power of electric vehicles For time period The total predicted power of electric vehicles, For time period Real-time electricity prices, including purchase price and sales price. For time period The planned electricity price for the day;

[0448] The adaptive Wasserstein fuzzy set building unit is used to construct adaptive Wasserstein fuzzy sets based on the global uncertainty vector:

[0449] ;

[0450] ;

[0451] ;

[0452] In the above formula, For adaptive Wasserstein fuzzy sets, Global uncertainty vector The true probability distribution To support the collection The set of all probability distributions above, the support set Global uncertainty vector The set of all possible values. for and The 1-Wasserstein distance between two distributions is used to measure the difference between them. Distribution based on historical experience, For adaptive Wasserstein fuzzy set radius, This provides an estimate of the local kernel density of the current predicted scenario within historical scenarios. Based on the radius, For adaptive adjustment coefficients, For the number of historical scene samples, For kernel function, For the first A sample of historical scenes;

[0453] The day-ahead and intraday model building units are used to construct day-ahead scheduling models and intraday rolling optimization models for photovoltaic-storage-charging stations based on adaptive Wasserstein fuzzy sets.

[0454] In the day-ahead and intraday model construction unit, the day-ahead scheduling model of the photovoltaic-storage charging station is a two-stage sub-Bruker optimization model. The first stage refers to the day-ahead baseline plan that must be made before the uncertainty is realized, and the second stage refers to the real-time adjustment decision made based on the day-ahead baseline plan after the uncertainty is realized.

[0455] The optimization model for the first stage of the Bruker bar is described in Example 1;

[0456] The second-stage optimization model for the Brussels bar is described in Example 1.

[0457] In the day-ahead and intraday model construction units, the intraday rolling optimization model is described in Example 1.

[0458] The real-time layer optimization module uses the real-time layer optimization model described in Example 1.

Claims

1. A multi-scale optimization method for photovoltaic-storage-charging stations considering multiple uncertainties, characterized in that, The method includes: S1. Considering the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices, construct a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage charging stations. S2. Solve the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station; S3. Input the day-ahead scheduling plan of the photovoltaic-storage charging station into the intraday rolling optimization model to obtain the optimal decision for each intraday rolling optimization period; S4. Based on the optimal decision for each rolling optimization period within the day, a real-time layer optimization model is constructed to obtain the real-time control commands for the photovoltaic-storage charging station system.

2. The multi-scale optimization method for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 1, characterized in that, S2 includes: S21. Construct the global uncertainty vector of the photovoltaic-storage-charging station system: ； ； ； ； ； In the above formula, This is the global uncertainty vector. For time period The uncertainty vector, This represents the total number of time periods within the scheduling cycle. For time period The error in photovoltaic power output prediction For time period Electric vehicle charging power prediction error, For time period The fluctuation deviation between the real-time electricity price and the planned electricity price before the day, For time period The actual power of photovoltaic power generation at that time For time period The predicted value of photovoltaic power generation, For time period The actual total power of electric vehicles For time period The total predicted power of electric vehicles, For time period Real-time electricity prices, including purchase price and sales price. For time period The planned electricity price for the day; S22. Construct an adaptive Wasserstein fuzzy set based on the global uncertainty vector: ； ； ； In the above formula, For adaptive Wasserstein fuzzy sets, Global uncertainty vector The true probability distribution To support the collection The set of all probability distributions above, the support set Global uncertainty vector The set of all possible values. for and The 1-Wasserstein distance between two distributions is used to measure the difference between them. Distribution based on historical experience, For adaptive Wasserstein fuzzy set radius, This provides an estimate of the local kernel density of the current predicted scenario within historical scenarios. Based on the radius, For adaptive adjustment coefficient, For the number of historical scene samples, For kernel function, For the first A sample of historical scenes; S23. Based on adaptive Wasserstein fuzzy sets, construct a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage-charging stations.

3. The multi-scale optimization method for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 2, characterized in that, In S23, the day-ahead scheduling model of the photovoltaic-storage charging station is a two-stage sub-Bruker optimization model. The first stage refers to the day-ahead baseline plan that must be made before the uncertainty is realized, and the second stage refers to the real-time adjustment decision made based on the day-ahead baseline plan after the uncertainty is realized. The first-stage optimization model for the split-bar is as follows: ； ； ； ； In the above formula, for feasible domain, These are the decision variables for the first phase of the recent scheduling. The first-stage cost is calculated based on the planned values ​​of the decision variables. For grid interaction costs, For energy storage depreciation costs, This represents the total number of time periods within the scheduling cycle. , Time periods Electricity purchase price and electricity sales price , These are the time periods planned for the current day. The power purchased from the grid and the power sold to the grid at that time. For time interval step, The energy storage depreciation cost per unit charge / discharge capacity. , These are the time periods planned for the current day. The charging power and discharging power of the energy storage device; The decision variables for the first stage are: ； In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy; The constraints of the first-stage sub-Bruker optimization model include system power balance constraints, photovoltaic constraints, energy storage system constraints, electric vehicle charging load constraints, and grid interaction constraints. The system power balance constraint is: ； In the above formula, For time period The actual power of photovoltaic power generation at that time For time period The basic load of the charging station For time period Total electric vehicle charging load; Photovoltaic constraints are: ； In the above formula, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic output prediction is one of the uncertainties that the system needs to deal with; The constraints of the energy storage system are: ； ； ； ； ； ； ； In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy. , These refer to the charging efficiency and discharging efficiency of energy storage, respectively. For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The rated power of the energy storage , These are binary variables representing energy storage charging and discharging, respectively. The electric vehicle charging load constraint is: ； ； ； ； In the above formula, , The first The time it takes for an electric vehicle to arrive at and leave the photovoltaic-storage charging station. For time period Time assigned to the first The charging power of an electric vehicle For the first The charging demand of a vehicle. For the first The maximum power of a charging station for electric vehicles. For time period The total power of the electric vehicle at that time For electric vehicle collection; The power grid interaction constraints are: ； ； ； In the above formula, , These are binary variables for purchasing and selling electricity, respectively. This represents the maximum power of the power grid.

4. The multi-scale optimization method for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 3, characterized in that, The second-stage optimization model for the split-bar is as follows: ； ； ； ； ； In the above formula, For the true probability distribution, For adaptive Wasserstein fuzzy sets, The value function for the second stage represents the value given the decision made in the first stage. Given the realization of uncertainty, finding the minimum cost required for optimal adjustment is itself an optimization problem. This represents the total number of time periods within the scheduling cycle. This is the cost coefficient for the second phase of adjustment. To adjust the amount, To account for the cost of energy storage battery degradation, , Time periods The charging power and discharging power of energy storage For time period Actual state of charge adjustment of energy storage It is a dynamic coefficient related to the SOC state. , These refer to the charging efficiency and discharging efficiency of energy storage, respectively. For time interval step, The investment cost per unit capacity of energy storage batteries, The rated capacity of the battery. For the battery at the reference depth of discharge The number of loops below, For time period depth of discharge, Battery life degradation index For time period The initial state of charge of the stored energy; The decision variables for the second stage are: ； In the above formula, For the decision variables of the second phase of the current scheduling, This is the global uncertainty vector. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Adjustment amount of electric vehicle charging load; The constraints of the second-stage sub-Bluerg optimization model include power balance constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints. The power balance constraint is: ； In the above formula, For time period Adjustment amount of power purchased from the grid. For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic power output prediction For time period The basic load of the charging station For time period Total predicted electric vehicle power, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of power sold to the grid; The actual charge / discharge constraint is: ； ； ； ； ； In the above formula, The rated power of the energy storage , These are binary variables representing the actual charging and discharging of the energy storage, respectively, characterizing the actual operating state. It is a sufficiently large constant; The actual constraints of the SOC are: ； ； ； ； In the above formula, For time period The actual state of charge of the energy storage at the beginning. , These refer to the charging efficiency and discharging efficiency of energy storage, respectively. For time period The actual charging power of energy storage For time interval step, For time period The actual discharge power of the energy storage For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The power grid interaction constraints are as follows: ； ； ； ； ； In the above formula, This is the maximum power of the power grid. , These are binary variables representing the actual electricity purchases and sales, respectively.

5. The multi-scale optimization method for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 2, characterized in that, In S23, the objective function of the intraday rolling optimization model is: ； ； ； ； In the above formula, Let be the decision variable, representing the time period determined in the intraday rolling optimization. The intraday planned value, Let be an optimized coverage window, representing the time set of the scheduling cycle for the intraday rolling optimization model. For operating costs, consistent with the current day's target, the calculation window uses the latest forecasted operating costs. For planning deviation penalties, This is a time-varying adaptive penalty coefficient, dynamically adjusted according to the uncertainty level of ultra-short-term forecasting. For time period The day-ahead scheduling plan is a known quantity. Used to measure the magnitude of deviation. For the true probability distribution, Represents a fuzzy set of intraday uncertainty. This is the second-order value function for the intraday phase, with the same meaning as the day-ahead scheduling problem, but the time range is limited to the intraday rolling window. For each scrolling optimization moment, For a fixed prediction window, Based on the rated penalty coefficient, This is the gain coefficient. For time period Uncertain mapping function , , These are the weighting coefficients for different degrees of uncertainty. For time period Photovoltaic power output, For time period Electric vehicle charging power, For time period Real-time electricity price; The constraint body of the intraday rolling optimization model includes power balance constraints, initial state constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints. The power balance constraint is: ； In the above formula, , These are the time periods in the intraday plan. The power purchased from the grid and the power sold to the grid at that time. For time period Adjustment amount of power purchased from the grid. For ultra-short-term photovoltaic power forecasting, For time period The error in photovoltaic power output prediction , These are the time periods in the intraday plan. The charging power and discharging power of energy storage. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period The basic load of the charging station For the latest ultra-short-term electric vehicle power forecast, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of electric vehicle charging load. For time period Adjustment amount of power sold to the grid; The initial state constraints are: ； In the above formula, For time period The state of charge, For time period Measured value of state of charge; The actual charge / discharge constraints, actual SOC constraints, and grid interaction constraints are the same as the second-stage constraints of the day-ahead scheduling model, and the planned and adjusted values ​​of all equipment must be satisfied.

6. The multi-scale optimization method for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 1, characterized in that, In S4, the objective function of the real-time layer optimization model is: ； ； ； In the above formula, The optimization goal for the real-time layer is... To predict the total number of steps in real time, , , These are all weighting coefficients for tracking error, respectively measuring the importance of tracking energy storage power, grid power, and SOC trajectory. For the current control moment, , , They are respectively in Time prediction The actual values ​​of energy storage at any given time, the actual power interacting with the grid, and the actual state of charge of the energy storage are the optimization variables for the real-time layer. , , They are obtained by intraday rolling optimization. The reference values ​​for energy storage at any given time, the reference values ​​for power interacting with the grid, and the reference values ​​for the state of charge of energy storage are known quantities. The weighting coefficient for the increment of the control quantity. To control the increase in quantity, For model credibility weights, This is the predicted output of the LSTM network for the system state. For the actual system output, The real-time system efficiency output by LSTM. To predict the time-domain step size in real time, The total energy that can be stored or provided under rated conditions. Compensation residuals for LSTM predictions; The constraints of the real-time layer optimization model include power constraints and equipment operation constraints.

7. A multi-scale optimization system for photovoltaic-storage-charging stations that considers multiple uncertainties, characterized in that: The system includes a model building module, a day-ahead scheduling model solving module, an intraday rolling optimization module, and a real-time layer optimization module; The model building module is used to consider the uncertainties of photovoltaic output, electric vehicle charging, and electricity prices to build a day-ahead scheduling model and an intraday rolling optimization model for photovoltaic-storage charging stations. The day-ahead scheduling model solving module is used to solve the day-ahead scheduling model to obtain the day-ahead scheduling plan for the photovoltaic-storage charging station; The intraday rolling optimization module is used to input the day-ahead scheduling plan of the photovoltaic-storage charging station into the intraday rolling optimization model to obtain the optimal decision for each rolling optimization period within the day; The real-time layer optimization module is used to construct a real-time layer optimization model based on the optimal decision of each rolling optimization period within the day, and obtain real-time control commands for the photovoltaic-storage charging station system.

8. The multi-scale optimization system for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 7, characterized in that, The model building module includes a global uncertainty vector building unit, an adaptive Wasserstein fuzzy set building unit, and a day-ahead and intraday model building unit. The global uncertainty vector construction unit is used to construct the global uncertainty vector of the photovoltaic-storage-charging station system. ； ； ； ； ； In the above formula, This is the global uncertainty vector. For time period The uncertainty vector, This represents the total number of time periods within the scheduling cycle. For time period The error in photovoltaic power output prediction For time period Electric vehicle charging power prediction error, For time period The fluctuation deviation between the real-time electricity price and the planned electricity price before the day, For time period The actual power of photovoltaic power generation at that time For time period The predicted value of photovoltaic power generation, For time period The actual total power of electric vehicles For time period The total predicted power of electric vehicles, For time period Real-time electricity prices, including purchase price and sales price. For time period The planned electricity price for the day; The adaptive Wasserstein fuzzy set building unit is used to construct adaptive Wasserstein fuzzy sets based on the global uncertainty vector: ； ； ； In the above formula, For adaptive Wasserstein fuzzy sets, Global uncertainty vector The true probability distribution To support the collection The set of all probability distributions above, the support set Global uncertainty vector The set of all possible values. for and The 1-Wasserstein distance between two distributions is used to measure the difference between them. Distribution based on historical experience, For adaptive Wasserstein fuzzy set radius, This provides an estimate of the local kernel density of the current predicted scenario within historical scenarios. Based on the radius, For adaptive adjustment coefficient, For the number of historical scene samples, For kernel function, For the first A sample of historical scenes; The day-ahead and intraday model building units are used to construct day-ahead scheduling models and intraday rolling optimization models for photovoltaic-storage-charging stations based on adaptive Wasserstein fuzzy sets.

9. The multi-scale optimization system for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 8, characterized in that, In the day-ahead and intraday model construction unit, the day-ahead scheduling model of the photovoltaic-storage charging station is a two-stage sub-Bruker optimization model. The first stage refers to the day-ahead baseline plan that must be made before the uncertainty is realized, and the second stage refers to the real-time adjustment decision made based on the day-ahead baseline plan after the uncertainty is realized. The first-stage optimization model for the split-bar is as follows: ； ； ； ； In the above formula, for feasible domain, These are the decision variables for the first phase of the recent scheduling. The first-stage cost is calculated based on the planned values ​​of the decision variables. For grid interaction costs, For energy storage depreciation costs, This represents the total number of time periods within the scheduling cycle. , Time periods Electricity purchase price and electricity sales price , These are the time periods planned for the current day. The power purchased from the grid and the power sold to the grid at that time. For time interval step, The energy storage depreciation cost per unit charge / discharge capacity. , These are the time periods planned for the current day. The charging power and discharging power of the energy storage device; The decision variables for the first stage are: ； In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy; The constraints of the first-stage sub-Bruker optimization model include system power balance constraints, photovoltaic constraints, energy storage system constraints, electric vehicle charging load constraints, and grid interaction constraints. The system power balance constraint is: ； In the above formula, For time period The actual power of photovoltaic power generation at that time For time period The basic load of the charging station For time period Total electric vehicle charging load; Photovoltaic constraints are: ； In the above formula, For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic output prediction is one of the uncertainties that the system needs to deal with; The constraints of the energy storage system are: ； ； ； ； ； ； ； In the above formula, For the planned time period in the previous day The initial state of charge of the stored energy. , These refer to the charging efficiency and discharging efficiency of energy storage, respectively. For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The rated power of the energy storage , These are binary variables representing energy storage charging and discharging, respectively. The electric vehicle charging load constraint is: ； ； ； ； In the above formula, , The first The time it takes for an electric vehicle to arrive at and leave the photovoltaic-storage charging station. For time period Time assigned to the first The charging power of an electric vehicle For the first The charging demand of a vehicle. For the first The maximum power of a charging station for electric vehicles. For time period The total power of the electric vehicle at that time For electric vehicle collection; The power grid interaction constraints are: ； ； ； In the above formula, , These are binary variables for purchasing and selling electricity, respectively. This represents the maximum power output of the power grid. The second-stage optimization model for the split-bar is as follows: ； ； ； ； ； In the above formula, For the true probability distribution, For adaptive Wasserstein fuzzy sets, The value function for the second stage represents the value given the decision made in the first stage. Given the realization of uncertainty, finding the minimum cost required for optimal adjustment is itself an optimization problem. This represents the total number of time periods within the scheduling cycle. This is the cost coefficient for the second phase of adjustment. To adjust the amount, To account for the cost of energy storage battery degradation, , Time periods The charging power and discharging power of energy storage For time period Actual state of charge adjustment of energy storage It is a dynamic coefficient related to the SOC state. , These refer to the charging efficiency and discharging efficiency of energy storage, respectively. For time interval step, The investment cost per unit capacity of energy storage batteries, The rated capacity of the battery. For the battery at the reference depth of discharge The number of loops below, For time period depth of discharge, Battery life degradation index For time period The initial state of charge of the stored energy; The decision variables for the second stage are: ； In the above formula, For the decision variables of the second phase of the current scheduling, This is the global uncertainty vector. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period Adjustment amount of electric vehicle charging load; The constraints of the second-stage sub-Bluerg optimization model include power balance constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints. The power balance constraint is: ； In the above formula, For time period Adjustment amount of power purchased from the grid. For time period The predicted value of photovoltaic power generation, For time period The error in photovoltaic power output prediction For time period The basic load of the charging station For time period Total predicted electric vehicle power, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of power sold to the grid; The actual charge / discharge constraint is: ； ； ； ； ； In the above formula, The rated power of the energy storage , These are binary variables representing the actual charging and discharging of the energy storage, respectively, characterizing the actual operating state. It is a sufficiently large constant; The actual constraints of the SOC are: ； ； ； ； In the above formula, For time period The actual state of charge of the energy storage at the beginning. , These refer to the charging efficiency and discharging efficiency of energy storage, respectively. For time period The actual charging power of energy storage For time interval step, For time period The actual discharge power of the energy storage For the rated capacity of energy storage, , These are the lower and upper limits of the energy storage state of charge, respectively. The power grid interaction constraints are as follows: ； ； ； ； ； In the above formula, This is the maximum power of the power grid. , These are binary variables representing the actual electricity purchased and sold, respectively. In the aforementioned day-ahead and intraday model construction units, the objective function of the intraday rolling optimization model is: ； ； ； ； In the above formula, Let be the decision variable, representing the time period determined in the intraday rolling optimization. The intraday planned value, Let be an optimized coverage window, representing the time set of the scheduling cycle for the intraday rolling optimization model. For operating costs, consistent with the current day's target, the calculation window uses the latest forecasted operating costs. For planning deviation penalties, This is a time-varying adaptive penalty coefficient, dynamically adjusted according to the uncertainty level of ultra-short-term forecasting. For time period The day-ahead scheduling plan is a known quantity. Used to measure the magnitude of deviation. For the true probability distribution, Represents a fuzzy set of intraday uncertainty. This is the second-order value function for the intraday phase, with the same meaning as the day-ahead scheduling problem, but the time range is limited to the intraday rolling window. For each scrolling optimization moment, For a fixed prediction window, Based on the rated penalty coefficient, This is the gain coefficient. For time period Uncertain mapping function , , These are the weighting coefficients for different degrees of uncertainty. For time period Photovoltaic power output, For time period Electric vehicle charging power, For time period Real-time electricity price; The constraint body of the intraday rolling optimization model includes power balance constraints, initial state constraints, actual charge and discharge constraints, actual SOC constraints, and grid interaction constraints. The power balance constraint is: ； In the above formula, , These are the time periods in the intraday plan. The power purchased from the grid and the power sold to the grid at that time. For time period Adjustment amount of power purchased from the grid. For ultra-short-term photovoltaic power forecasting, For time period The error in photovoltaic power output prediction , These are the time periods in the intraday plan. The charging power and discharging power of energy storage. , Time periods The adjustment amount of charging power and discharging power for energy storage. For time period The basic load of the charging station For the latest ultra-short-term electric vehicle power forecast, For time period Electric vehicle charging power prediction error, For time period Adjustment amount of electric vehicle charging load. For time period Adjustment amount of power sold to the grid; The initial state constraints are: ； In the above formula, For time period The state of charge, For time period Measured value of state of charge; The actual charge / discharge constraints, actual SOC constraints, and grid interaction constraints are the same as the second-stage constraints of the day-ahead scheduling model, and the planned and adjusted values ​​of all equipment must be satisfied.

10. The multi-scale optimization system for photovoltaic-storage-charging stations considering multiple uncertainties according to claim 7, characterized in that, In the real-time layer optimization module, the objective function of the real-time layer optimization model is: ； ； ； In the above formula, The optimization goal for the real-time layer is... To predict the total number of steps in real time, , , These are all weighting coefficients for tracking error, respectively measuring the importance of tracking energy storage power, grid power, and SOC trajectory. For the current control moment, , , They are respectively in Time prediction The actual values ​​of energy storage at any given time, the actual power interacting with the grid, and the actual state of charge of the energy storage are the optimization variables for the real-time layer. , , They are obtained by intraday rolling optimization. The reference values ​​for energy storage at any given time, the reference values ​​for power interacting with the grid, and the reference values ​​for the state of charge of energy storage are known quantities. The weighting coefficient for the increment of the control quantity. To control the increase in quantity, For model credibility weights, This is the predicted output of the LSTM network for the system state. For the actual system output, The real-time system efficiency output by LSTM. To predict the time-domain step size in real time, The total energy that can be stored or provided under rated conditions. Compensation residuals for LSTM predictions; The constraints of the real-time layer optimization model include power constraints and equipment operation constraints.

Images